<![CDATA[This study presents COMPASS, a comparative evaluation of ten Machine Learning algorithms for DDoS attack detection using the AISED Dataset on Cloud DDoS Attacks. Feature selection was performed using SelectKBest with ANOVA F-Value, evaluating model performance across varying feature dimensions (K = 10, 15, 20, 25). Experimental results demonstrate that ensemble-based methods, particularly Random Forest, Gradient Boosting, and AdaBoost, achieve near-theoretical maximum AUC scores (>0.998) while maintaining fast training times (<0.1 seconds). K-Nearest Neighbors (KNN) also exhibits robust performance (AUC > 0.98) with minimal computational cost. In contrast, Support Vector Machine (SVM) and Quadratic Discriminant Analysis (QDA) show relatively lower accuracy (AUC > 0.85) and suffer from high computational complexity, with SVM requiring up to 572 seconds to train at K=25. These findings highlight the critical trade-off between classification accuracy and computational efficiency in selecting optimal models for real-time DDoS detection systems. As future work, we propose deploying a lightweight version of COMPASS on edge computing devices and integrating it into federated learning frameworks to enable collaborative, privacy preserving model training.]]>
Copyrights © 2025
